33 Claude Skills for Design Thinking That Ships

What these skills are

A whole design practice, as installable Claude skills. The pack covers the whole arc of a design project: framing the challenge, preparing the research, making sense of what real people told you, defining the problem, shaping the ideas, testing them with real users, and landing the whole thing in front of the people who decide.

One rule runs through all of it: Claude prepares and synthesizes, humans meet the users. No skill here replaces a single conversation with a real customer. Every persona, insight and finding in this pack is built from research with real people, and the pack will tell you when it is time to go and talk to them. What it takes off your plate is the paper trail: the guides, the debriefs, the synthesis, the cards, the test plans, the playback.

Mechanically, each skill is one folder with a SKILL.md file. The 33 are grouped into seven phases, and each one gives full value on its own: run one, run a phase, or chain the lot. They work from one Claude Project per challenge, so the brief, the themes and the concepts stay in one findable place and the work accumulates instead of scattering.

Download all 33 skills. One zip: ready-to-install skill zips, readable SKILL.md files, and the five blank templates. Free, no signup. The same 33 skills are open on GitHub: polar-bear-org/claude-skills.

The 33 skills you get

The set follows the arc of a real project, in seven phases: frame, prepare, make sense, define, shape, test, land.

1 · Frame the challenge

1. Design Challenge Framer

Use when: A project is starting and the problem is still a feeling rather than a statement

Output: A one to two page challenge brief: context, the problem statement, scope in and out, constraints, success criteria, open questions. Nearly every other skill in the pack reads it first

2. Stakeholder Mapper

Use when: The project is starting and it is unclear who can kill it

Output: The grid as a table, an engagement note per key player, and the missing-voices list — who is affected but has nobody in the room. One page, two at most

3. Assumption Mapper

Use when: The team is confident and nobody can say where the confidence comes from

Output: The full know/believe sort, the top five risks ranked with a one-sentence kill scenario each, and the cheapest test per risk with rough effort. Under two pages

4. Desk Research Digest

Use when: A discovery is starting and old reports, analytics and support logs are sitting unread

Output: Established findings with sources, suggestive findings with their caveats, contradictions, and the questions only primary research can answer. Every claim traceable to a named source

2 · Prepare the research

5. Interview Guide Designer

Use when: Interviews are scheduled and the question list is a blank page

Output: Intro script with the consent line, warm-up, timed arcs with probes, close. One to two pages, printable, opening with the reminder that the guide serves the conversation and not the other way around

6. Participant Screener Builder

Use when: Recruitment is starting and "eight users" has to become eight specific kinds of person

Output: The screener ready to paste into any form tool, the answer key showing which responses qualify and why, quotas, and disqualifiers

7. Survey Designer

Use when: Someone wants numbers from many users and the draft questions all start with "don't you agree"

Output: Goal statement, the questions with types and answer options, the analysis plan per question, pilot instructions, and a stated length cap with an estimated completion time

8. Observation Plan Builder

Use when: The team suspects that what users say and what users do are two different stories

Output: Focus list, capture sheet, visit logistics, consent script, and the same-day debrief checklist. Two pages, field-printable

9. Diary Study Designer

Use when: The question involves habits, moods or moments nobody can recall accurately afterwards

Output: Research questions, trigger design, the prompts verbatim, the participant onboarding script, an honest burden calculation, duration and sample, the incentive note, and the follow-up interview plan

10. Research Kickoff Brief

Use when: Methods and guides exist but nobody has agreed on paper who is doing what by when

Output: One page: questions, methods with rationale, sample, timeline, roles, and a "decisions this research feeds" line at the top so nobody forgets why it is happening

3 · Make sense of it

11. Interview Debriefer

Use when: An interview just ended and the notes are a wall of raw text

Output: One page per session: participant descriptor with no real name, quotes by territory, observed behaviors, surprises, follow-ups, and the interpretation section marked as interpretation. These files are what the synthesizer reads

12. Transcript Synthesizer

Use when: A stack of sessions is done and the team needs themes instead of nine separate stories

Output: Each theme with its definition, count, supporting quotes with participant tags, counter-examples and a confidence note; then single-voice observations, tensions, and open questions

13. Survey Analyzer

Use when: Response data exists and someone is about to average their way to a wrong conclusion

Output: The evidence base first — n, response rate, fielding dates, known biases — then findings with distributions and segment cuts, then the section on what this data cannot tell you

14. Insight Writer

Use when: Themes exist and the team needs the sentences that make a room lean forward

Output: Five to nine insight statements, each with its evidence trail and a confidence note, ranked, with the mere confirmations listed separately at the end

15. Empathy Map Builder

Use when: Research exists and the team wants the says, thinks, does, feels view of a segment

Output: One page per map: the four quadrants plus pains and gains, every entry sourced or explicitly marked as inference, an evidence footer, and the gap list

4 · Define the problem

16. Persona Designer

Use when: Research data exists and the team needs people-shaped summaries of who they are designing for

Output: Three to five personas, each under a page, each with its evidence footer and at least two verbatim quotes from real participants, plus a summary table: persona, core goal, sample size behind it

17. JTBD Writer

Use when: The team keeps describing features and needs to describe the progress people are trying to make

Output: Job statements with their functional, emotional and social layers and evidence tags, the outcome ranking (or the honest note that ranking needs more data), and the opportunity flags

18. Journey Builder

Use when: The team needs to see the whole experience end to end instead of their own touchpoint

Output: The structured map — stages as columns; actions, touchpoints with owners, thoughts, feelings, pains, moments of truth and opportunities as rows — plus a narrative a stakeholder reads in three minutes. Unresearched stages are marked, not smoothed over

19. Five Whys Runner

Use when: A pain point is well documented and the team is about to solve its symptom

Output: The causal chain with evidence at each step, branches shown, unverified rungs marked with their follow-up research questions, and the recommended solving level with reasoning

20. Problem Statement Writer

Use when: The research is synthesized and the team must choose which problem gets solved

Output: Three to five candidate POVs, each with its evidence trail, traits check and the trade-off that framing implies; a comparison table; and a recommendation with reasoning

21. HMW Generator

Use when: A problem statement exists and an ideation session needs its launching questions

Output: Eight to fifteen clustered How Might We questions, each traceable to its POV or insight, with the rejected-and-rewritten examples shown so the team learns the policing move. One page, printable for the room

5 · Shape the ideas

22. Idea Expander

Use when: Everyone keeps suggesting the same three obvious solutions

Output: Twenty or more directions grouped by distance, each tagged with the lens that produced it and the insight or HMW it serves. Labeled as desk divergence: raw input for the team's judgment, not a ranked list

23. Concept Card Writer

Use when: Ideation produced winners and they need a fair, comparable write-up before prioritization

Output: One page per concept in an identical format, plus an index table: concept, who it serves, core bet, first test. These cards are exactly what the matrix and the DFV check read

24. Value Impact Matrix

Use when: A set of concepts exists and the team must choose where the effort goes

Output: The 2x2 as a table with per-concept reasoning, the calibration example and axis definitions, the now/next/later read with owners where known, and the parked list with revival conditions

25. DFV Checker

Use when: Concepts are about to get investment and nobody has asked the three hard questions on paper

Output: Per concept: desirability, feasibility and viability with evidence, verdict and confidence; the weakest leg named; the strengthening prescription. No traffic-light greens without cited evidence behind them

26. Role Play Designer

Use when: A service idea exists only as boxes on a slide and nobody has lived it yet

Output: Scenario, one role card per player, the beat script, props list, observer sheet and debrief questions. Ready to run in 60 to 90 minutes with four to eight people

27. Storyboard Writer

Use when: A concept needs to be felt as an experience before anyone commits to building it

Output: Six to eight numbered frames, each with scene description, caption and emotional beat, plus a two-line story summary. Ready to hand an illustrator or sketch from — stick figures are fully sufficient

6 · Test with real people

28. Riskiest Assumption Finder

Use when: A concept is heading toward a build and nobody has named the bet underneath it

Output: Per concept: the assumption stack ranked with reasoning, the first-test assumption flagged, and its cheapest honest test with rough effort and the evidence it would produce

29. Prototype Planner

Use when: An assumption needs testing and the team's instinct is to build the product to find out

Output: The question, the chosen method with rationale, build scope with a day cap, the evidence bar committed to before the sessions, the session plan with real-user recruitment, and the cannot-answer list. One page

30. Test Script Designer

Use when: Sessions are booked and the moderator needs words that produce behavior instead of compliments

Output: Intro and consent script, tasks with success criteria, think-aloud prompts, the question bank in order, an observer sheet, and timing for the slot. Ready to moderate from

31. Test Debrief Synthesizer

Use when: The sessions are done and the team is about to remember them more fondly than they went

Output: The evidence base, behavior-versus-opinion findings with counts and quotes, per-assumption verdicts held against the bar set before the data, the recommendation with reasoning, and open questions

7 · Land it

32. Insight Playback Builder

Use when: The work is done and the people who were not in the room need to believe what the team now knows

Output: The narrative playback, answer-first, with the evidence base stated up front and the what-we-did-not-find section in the main body, optionally generated out as a deck. Every claim traces to a project artifact

33. Recommendation Writer

Use when: One page has to carry the decision to someone who was never in a session

Output: One page: the recommendation, what it rests on, what is still assumed, the first two weeks, and the walk-away condition. Jargon-free, no appendix required to parse it

How a project flows through the pack

One pinned chat per challenge carries the whole arc. design-challenge-framer writes the brief nearly every other skill reads first; the research skills turn it into guides, screeners and a kickoff one-pager; interview-debriefer files each session the day it happens and transcript-synthesizer finds the patterns across them. From there insight-writer and problem-statement-writer decide what is worth solving, hmw-generator hands the room its questions, the concept skills make the ideas comparable, and the test skills send you back out to real users before anything gets built. insight-playback-builder and recommendation-writer close it for the people who decide.

Five blank templates ship baked into the skills that fill them in — challenge-brief, research-kickoff, interview-debrief, journey-map and concept-card — so the shape is on hand with no project setup. The same blanks sit in the pack's templates folder if you would rather use them on their own.

The pack prepares and synthesizes; it does not run the room. When a skill says "take this into a session", the session itself lives in the Workshop Pack.

Setup guide

1. Download the pack. One zip: an install folder with 33 ready-to-upload skill zips, a skills folder with the same skills as readable files, and the five blank templates.
2. Install your skills. In Claude Code, add the marketplace and install design-thinking-pack, and all 33 load at once. In Claude, turn on code execution in Settings, Capabilities, then go to Customize, Skills and upload one zip per skill from the install folder. Prefer working from files? Add the SKILL.md files to your Project knowledge instead; it works, just less cleanly.
3. Create a Project per challenge. Make one Project named after the challenge ("DT · Acme discovery") and start one pinned chat: that chat is where the project lives. Skills save their outputs as files named after the project, and later skills read the earlier artifacts by those names, so the work accumulates instead of scattering. Then write "run design-challenge-framer". It will ask what you are actually trying to change for whom, and it will not accept a solution dressed as a problem.

Where to start

The quality bar

Every skill in the pack holds the same standard, the one we hold when we run discovery ourselves:

• Claude prepares and synthesizes, humans meet the users: no skill runs an interview, and none invents one
• AI-invented users presented as research is the one thing this pack never does: no persona, insight or journey without real research under it
• Evidence means real data from real people or real systems: team conviction, a sponsor's certainty and a competitor's feature list move nothing out of the believe column
• Raw words stay raw: quotes are never sharpened, merged, or blended into a composite voice
• Silence is labeled silence: unresearched journey stages, thin themes and small samples say so instead of being smoothed over
• No solution smuggled into the question, not in the problem statement and not in the How Might We, however much the sponsor likes that solution
• The people affected go on the stakeholder map even when nobody in the room represents them
• The recommendation never outruns the evidence: a well-evidenced "not this" is a project that succeeded at its actual job

Who made this

Polar Bear is a people ops consultancy for human-size teams (20 to 200 people). Built by ex-McKinsey founders with a dream to make AI work for People, not instead of them. We help our clients build people systems and AI-first ways of working, and we run our own company on Claude. This pack is the free, self-serve version of how we work.

The pack carries one design project at a time. When you want your whole team working this way, AI carrying the overhead so people do the thinking, across discovery, workshops, and everyday work, that's what we build with clients.

Meet Pauline. Sitting on a fuzzy challenge, or a stack of interviews nobody has read? Bring the project and we will find where it actually starts. Book a 30-minute call · Pauline on LinkedIn